How to use the Ichimoku Kinko Hyo Signals agent skill

What the ichimoku skill makes the agent do

An agent skill is a folder of instructions and optional scripts that an AI agent can load when a task matches it. This one is named ichimoku and comes from the HKUDS/Vibe-Trading repository, under the MIT licence. It describes the five Ichimoku lines and the rules for turning them into signals.

The five lines are the Conversion line (Tenkan-sen), Base line (Kijun-sen), Leading Span A and B (which form the cloud, or Kumo), and the Lagging Span (Chikou). Signals are triggered only on Tenkan/Kijun crossover events and then filtered by cloud position and cloud direction.

• Tenkan-sen: (9-period highest high + 9-period lowest low) / 2, short-term trend
• Kijun-sen: (26-period highest high + 26-period lowest low) / 2, medium-term trend
• Senkou Span A: (Tenkan + Kijun) / 2, shifted forward 26 periods, upper cloud boundary
• Senkou Span B: (52-period highest high + 52-period lowest low) / 2, shifted forward 26 periods, lower cloud boundary
• Chikou Span: closing price shifted backward 26 periods, trend confirmation

Install the ichimoku skill

Option one is a personal install. Download the ZIP from the skill page, unzip it, and place the ichimoku folder in ~/.claude/skills/. Option two is a project install: put the folder in the .claude/skills/ directory of the project you are working in, so only that project uses it.

If you use the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

The skill's SKILL.md lists its Python dependencies. Install them with the command below before running any code.

• Personal: ~/.claude/skills/ichimoku
• Project: .claude/skills/ichimoku inside your project folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

How to trigger the skill with natural requests

You do not need a special command. Describe the task in plain language that mentions Ichimoku, the lines, or the signal you want, and attach or point to your OHLC price data (open, high, low, close). The agent should load the skill when the request matches its description.

Be specific about your data and the output you want. State which timeframe the candles are in, and whether you want the latest signal or a signal series for backtesting.

• Ask for the signal: "Using the ichimoku skill, what is the signal on the last candle of this OHLC file?"
• Ask for a calculation: "Calculate Tenkan-sen, Kijun-sen and both Senkou spans for my EURUSD daily data."
• Ask for a backtest: "Use the ichimoku skill to generate signals on this price history and count the long and short signals."
• Ask for an explanation: "Explain why the ichimoku skill says stand aside on this candle."

Understand the signal rules

A signal is produced only when a Tenkan/Kijun crossover happens. A strong buy needs three conditions to agree: a bullish TK cross, price above the cloud, and a bullish cloud where Span A is greater than Span B. A strong sell needs the mirror conditions: a bearish TK cross, price below the cloud, and Span A below Span B. Any other case means stand aside, which is returned as 0.

The engine needs at least 78 candles before it produces valid signals, because the 52-period calculation plus the 26-period displacement requires that history. The default parameters are 9, 26, 52 and 26, and the skill documents them so you can see what can be changed.

• 1 = long (strong buy)
• -1 = short (strong sell)
• 0 = stand aside (all other cases)
• Warm-up: 78 candles minimum (52 + 26)

Check what the skill needs and what it runs

You need Python, the pandas and numpy libraries, and the requests library listed in SKILL.md. The skill does not fetch market data itself: you supply the OHLC price data. The source's excerpt of the example code does not use requests, so it may be installed but unused in what is shown.

The package includes an example_signal_engine.py script, so the skill runs scripts. The listing states that it does not use the network.

• Required: Python, pandas, numpy, requests (install list in SKILL.md)
• Your input: OHLC price data supplied by you
• Runs scripts: yes (example_signal_engine.py)
• Uses network: no

Safety notes before you rely on the output

Read example_signal_engine.py before running it on data you care about, since the skill runs code on your machine. Keep it inside a folder you trust and review any code the agent writes from it.

Treat the output as a rule-based calculation, not a forecast. The skill does not validate its signals in the source, and it does not model costs, slippage, or data gaps. Test any rule independently before using it with real money.

• Review scripts before running them
• Check your price data for gaps and missing candles
• Test with a backtest and out-of-sample data first
• Do not treat a signal as financial advice

Tips for better results

Give the agent clean, ordered candles at a single timeframe, with enough history to pass the 78-candle warm-up. If you change the default periods, say so in your request so the results are interpretable.

The reference notes in the skill also describe medium and weak crossovers depending on where the cross happens relative to the cloud. The SKILL.md signal rules are the strong buy and strong sell cases, so ask the agent which logic it applied if the result is unclear.

• Name the timeframe and the number of candles you supplied
• Keep default periods unless you have a reason to change them
• Ask the agent to show the five line values next to each signal